Encode cover-hurdle data for the two-Laplace fit — encode_cover

Splits y into a binomial occurrence indicator and a positive-cover subset, builds design matrices for each arm using the fixed-effects part of formula, and extracts any structured terms it carried (an areal icar()/bym2() spatial field, plus temporal() / re() blocks). The formula is parsed against the NA-dropped observations so the structured index codes align with both arms.

Usage

encode_cover_hurdle(
  formula,
  data,
  y,
  positive = c("lognormal", "lognormal_trunc", "beta", "beta_oi", "ordinal", "gaussian"),
  breaks = NULL,
  autoscale = TRUE,
  presence_formula = NULL,
  positive_formula = NULL
)

Arguments

formula: State-process formula (no LHS); used for both occurrence and positive-cover arms.
data: Data frame with nrow(data) == length(y).
y: Length-N numeric vector of cover in [0, 1]. NAs are dropped from both arms (treated as missing, not as zero cover).
positive: Positive-arm likelihood: one of "lognormal", "lognormal_trunc", "beta", "beta_oi", "ordinal", "gaussian".
breaks: Ordinal cut points, required when positive = "ordinal" and ignored otherwise.
autoscale: Logical (default TRUE); rescale the positive arm away from the boundary so the Laplace engine never sees an exact 0 or 1.
presence_formula: Optional formula overriding formula on the occurrence arm. NULL uses formula for both arms.
positive_formula: Optional formula overriding formula on the positive-cover arm. NULL uses formula for both arms.

Value

A list with: occ_data, pos_data, spatial_spec (a tulpa_spatial built from the unweighted areal formula term, or NULL), trend (per-observation weight + label from a weighted areal term, or NULL), temporal and re (structured terms from the formula, or NULL), N, idx_pos (row indices of the positive subset within data), formula (the fixed-effects formula), positive.

Details

For positive = "lognormal" the positive arm's response is log(y[occur == 1]). For positive = "beta" it is y[occur == 1] on the natural (0, 1) scale; an additional eps-clip is applied so the Laplace engine does not see exact 1's introduced upstream.